How to Keep AI-Generated Content Sounding Human

What We Need to Grow, Episode 023: Signs of Humanity. A Founder Talk conversation with Matt Blumberg, CEO of Markup AI. Cleaned for readability; the words are the speakers' own.

Caleb Pedosiuk: Matt Blumberg, there are so many things in your experience that I have questions about. Everything from, going back, OG dot com boom. You were already building stuff back in that era. So you've got deep roots in the forest of what's going on right now. Right now you're working on a project right at the pulse of one of the most important things going on with gen AI and everybody integrating new tools, creating content, with Markup AI. But you have built and scaled other companies, you've raised money, you've done things at the enterprise level, you've run a couple of podcasts. I would love to hear about what's going on right now. But man, I have questions. So, what's going on right now?

Matt Blumberg: Let's go. Alright, what's going on right now. My current company is called Markup AI. And what we are trying to do at Markup AI is make it easy and safe and confident for people at companies to hit publish on their content.

The story is a simple one. For years, up until a couple of years ago, most content was created by humans. And that content had editors and a whole kind of content workflow and editorial process. By the way, our company has some legacy software products that helped writers and editors with that process, predating all of this. But large language models have completely and totally scrambled content workflow. And what we're seeing inside of large enterprises now is that this move to have AI do it all doesn't quite work. What it does is it turns everybody from a writer into an editor. They can be really good at prompts and tell Claude or Gemini or whatever Copilot what to write. And they get stuff back and it's good, it happened quickly, they didn't have to do all the writing themselves, but it's not 100 percent right. And businesses rely on their content being 100 percent right.

So that's where Markup AI comes in. We are a multi-surface, multi-agent, multiplayer solution that is really simple to implement. We have clients that get up and running self-serve within a few minutes. We basically live where their content lives. So we're either in a Chrome extension working on Google Docs, we're in a Microsoft Word extension, we have a really robust MCP server, we have a really well documented API. So we live where the content lives. Clients can determine their standards and rules in our application, so it's one brain and anyone in the enterprise can access it.

And the standards we help them evaluate their content against are the most important things they care about when they go to hit publish, where everyone has that moment of, am I ready? Did Claude get this right? So we look really closely at their tone, voice, terminology. Are they getting all that stuff right? We look really closely at compliance, regulatory compliance, which for different industries can mean nothing or it can mean everything. We look at AI search visibility, so what people are now calling AEO and GEO, but basically not just, can I be found by Google, but can I be found by Claude, can I be found by OpenAI, Perplexity, Grok. And then finally we help with accuracy and fact checking. So all those things, you put them together, it's a really complicated set of interlocking agents that are all in one engine we've created, to basically help get the AI content infrastructure up to speed. Basically we're trying to do for content what all these other companies have done for code over the last three years.

Caleb: And what percentage of content is word based versus other aspects of the brand?

Matt: Our system at the moment and our agents at the moment are really focused on text. It is still the overwhelming majority of the enterprise content that's online. And we are starting to do some multimodal work, we have some agents we're working on right now that get into images. I assume over time we'll get into video and other things as well. But right now it's text, and that is most of what corporate content is.

Caleb: The price of getting one thing wrong at the enterprise level, or in sensitive industries where you're dealing with healthcare, finance, the cost. One of the questions I had for this is, you're probably dealing with marketing leads or someone at the leadership level who's making decisions and overseeing. And I'm curious to know, people feel the noise of gen AI. Even LinkedIn this past week, with the whole "this sounds like AI slop" button, which is kind of funny for a tool that also supports generating AI content, helping AI write things. But I'm curious how the noise, people recognizing that maybe it's annoying, becomes a pain point. How you've been able to help leadership look and go, guys, this isn't just a convenience thing, there's actual liability here, that if these things carry on over a long enough period of time you're going to hit some kind of snag.

Matt: There is a real cost to having content out there that is not accurate. There is a real cost to having content out there that runs afoul of a law or regulation. And honestly there's a real cost in lost opportunity. And there is, I would say, maybe a less tangible cost, but there's a real cost to having content be off brand. The reality is, the way large language models work, the whole thing about the transformer and LLMs is that they're predictive. Which means everybody's content sounds the same. And when your content sounds like everyone else's content, how are you differentiated?

So whatever the vector is, companies that take their presence seriously online, which should be every company, need to make sure that their content works. There's the risk that you run afoul of a law. You can get sued by a regulator or a customer for millions, hundreds of millions of dollars, and there are lots of stories about that. And there is no legal defense that says, my AI did it, oops. If you are a regulated bank in the United Kingdom and you have to have marketing materials at the sixth grade reading level, and AI puts them out there at the twelfth grade reading level, you're exposed. If you have to have accurate toll-free phone numbers somewhere on your website for someone to call if they've got a question, you're exposed. If you're a pharmaceutical company and you write marketing copy that's not quite accurate to how the drug works, you've got a big problem on your hands.

Those are the regulatory ones, but the non-regulatory ones are equally important. Again, if your content doesn't sound human and authentic and like your brand, you are missing out on sales. If you're not discoverable by the way people look for things today, you are missing out on sales. And if your content is just plain inaccurate, you're going to piss off customers and prospects and you're missing sales. So either way, your content matters.

Caleb: And if I understand correctly, the foundation of Markup AI actually has compliance as a first order principle. Just looking at my notes here, Acrolinx. So you have this enterprise level understanding or expertise in compliance, before, as I understand it, you sort of expanded beyond that. Is that right?

Matt: The legacy company, Acrolinx, is actually 25 years old. I wouldn't say strictly compliance, but I would say complex enterprise content workflows, which means brand as well as compliance. And yes, that is what the company has been doing. It's been doing it since almost 2000. And we have deep roots doing that using older technologies like natural language processing and machine learning and machine translation. So Markup AI is the new name for the company, but it's also the new platform, which is native agentic AI that incorporates all of the best parts of our legacy system and also extends it into an ecosystem of LLMs, MCP, and the new rapid fire pace of innovation and development.

Caleb: Interesting that even this company, Acrolinx, being around for the dot com boom. I guess I go back to it, it's got a special place in my mind, because I remember the energy, the feel, coming out of school fresh out of studying marketing, and just the energy of the internet doing what it's saying, and the buzz of brick and mortar versus dot com, and so many visionaries or technologists envisioning what the prophetic future is, people hypothesizing. And it's funny to me that a lot of that hypothesizing was accurate. The language being used to describe it, or the scenarios, maybe not exactly, kind of clunky. But the timeline for a lot of the change was not immediate. A lot of it did require user adoption, hardware infrastructure.

Matt: Don't minimize the importance of having hardware catch up to software capabilities. But absolutely, so it is interesting. Look, I've been the CEO here for a year and a half now, so I'm the steward of the brand at the moment, and my team and I are reinventing the company in the current world, the current universe. There are lots of legacy software businesses that are going to have real problems at the moment, because they're not reinventing, or because they are taking a legacy platform and sprinkling in a little LLM here and a little LLM there and pretending like they're actually native AI, which they're not. We took an approach of start from scratch. We're still maintaining our old business, but the new business is done, the platform is up and running, it's solving problems for new clients every day, and we're migrating our legacy customers over to the new platform. Companies are handling the disruption from AI in lots of different ways, and that's how we're approaching it here, sort of a clean slate, but with knowledge and a lot of customers about how this stuff works in the real world.

Caleb: How are you preparing for the agent to agent component of things? A bit of context for that. One of the things that was an aha moment for me a couple of months back, as we sort of put the centre of the dartboard for us, the target of all the places you can be focused within a broader marketing category, we looked at showing up in AI search, AI visibility, as the centre, because if you can hit that, all these other things under the surface are working, are in proper order. So that's the focus. But within that, I realized the machine layer, optimizing your content, creating content for the machine layer, is crucial, and it actually serves the human layer better when you've done that properly.

Matt: Ain't that the truth. Sorry, I didn't mean to interrupt your question.

Caleb: Ain't that the truth. I think recognizing that, for example, a lot of creative teams, and we were much more heavily indexed when we started back in 2012 on building brand campaigns, visuals, the traditional understanding of brand. And then you quickly realize, if your brand isn't seen, or your campaign isn't seen by the right person at the right time, it's like it doesn't exist. So you had to go to the opposite side and understand pixels, tags, tracking, funnels. And then when you have them both, you sort of have this relationship of two hemispheres that can work together.

But it's interesting now, because there are campaigns that go viral, a lot of really intelligent, creative, strategic thinking that goes into that. A lot of our focus is really looking and saying, what are the machines looking for to understand if a company, if a brand, is solving a meaningful problem, and if they do it well? Because if they do, and one of these models will confidently cite them as an answer to someone's problem, it's magic.

Matt: No, it is. And I think the thing that's so funny about this, hey, machine to machine, is it readable, the agents are talking to the agents. One of the things that is both a little confusing but also a little comforting about that is, there is a human on each side of that equation. And it might be, depending on how you draw the schema out, it's like a layered sandwich of human and machine. So yes, you need things to be machine readable. But one of the things the machines are looking for are signs of humanity. And so if you don't have authentic voice and content that's accurate and that makes sense, it doesn't matter if it's structured correctly, it's never going to float to the top of the list.

So I think it's a little different than traditional SEO. And I'm not an expert in SEO, but it's a little bit different than, oh, we got the meta tags right. And it's also a lot harder to track. It's one thing, you can run the same search in Google three times and you kind of get the same answer. But three different people are going to run a similar query on AI engines and probably get three pretty different answers, depending on the exact wording of what they put in and what their chat history is with that particular engine. So all these companies that say, oh, we've solved AI search, I'm not sure they've exactly figured out how to do all the measurement and monitoring on the back end. But it is clear that there are certain rules and certain things the engines are looking for in order to pick you up, and in order to realize that you're legit. And from everything I've learned, and again, I'm not an expert in this, a lot of it is still based on Google and the power of OG search.

Caleb: For some of these larger systems looking to retool, and certainly for Google, the size of entire departments is so large, they're the size of some industries. That's one of the things we've noticed specifically with optimizing campaigns, whether it's for paid campaigns or on the SEO side. What you're hearing from one department in terms of recommendation versus another. I think it was Google, within about ten days, two different departments came out with different official positions on key factors in AI visibility. This was probably about a month and a half ago. So you can see that they're also retooling.

And it seems more logical. For example, when you have noticed people game a system, like with Reddit, with organic growth and comments in Reddit, and how some teams are sort of black hatting it in a way, and you find these grey areas, and then the models adjust. So I think from our perspective it has more to do with building a rich soil of content around the ecosystem of a brand, with as much of the real world's authentic aspect of engagement with customers, talking about product.

Matt: For sure. And by the way, that's no different than Google and page rank. I mean, it's the next generation of the whole thing. Page rank was interesting, it was the way to search for things, because it was about real citations. Well, that's what everyone's talking about now, real citations.

Caleb: So what do you find those conversations are like at the enterprise level? Do you find there's that sense of these larger ships in the ocean that take more time to turn, or is it more of a sense of not business as usual, and more eager to find ways to improve efficiencies across departments?

Matt: There's a great quote, and I can't remember who said it, which is, the future is already here, it's just not evenly distributed yet. So the answer is, we see everything. We see some clients that want us yesterday and are ready to go. We see others who are still taking a wait and see approach about AI in general. And we see everything in between.

We see a lot of companies that have a reflex on AI with everything, well, let me go build that myself. And I think that pendulum is swinging back a little bit towards buy and away from build in enterprises. But look, with AI you can build more stuff now than you've ever been able to build in the past. That doesn't necessarily mean you should, and it doesn't necessarily give you a properly evaluated, enterprise grade application. So there's a lot of noise around buy versus build. There's definitely the wait and see crowd. There's what I would classify as the normal very large corporate enterprise procurement motion, which is like, okay, we'll buy it, but we've got infosec, and what models are you using, and data, and things that maybe go a little bit overboard. But there's a really wide range. Like I said, we've had clients adopt our platform self-serve, get up and running and get value out of it in like five minutes. And we've had other clients where we're in a twelve month sales cycle. So it's all over the map right now.

And by the way, that's not a lot different, you're talking about the OG stuff, that's not a lot different than the way companies approached websites in 1995 and 1996. The website got invented and all of a sudden it was IT's responsibility, until it became clear that it's actually a marketing tool. And so then it moved from being IT to a partnership between marketing and IT. And now a lot of that stuff is really in the hands of users. And AI is going to follow that same curve.

Caleb: We've noticed some similarities, in terms of, typically when we have a conversation with someone we could help, they almost self-select. They're generally interested in enough information to get a sense, and then feel like they want to go away and do that work. And we found, just bless them, go for it, green lights, guys. Then you find others, and they recognize it's typically speed. A lot of the time they already have product market fit, they've got a strong team, and that team is already focused on what they're supposed to be doing, so they don't have people sitting around looking for something to do. They're not looking to add more things to that team to take internally. They're already focused. And at that point the cost isn't actually just a calculation of having you do that thing. There are second or third order consequences of impacting revenue, potential market share.

I have one question before my final question. My final question is, what is the most important thing you need to grow right now. But the quick question before that is, you've done a couple of podcasts, you've been creating content. I saw one project, and I don't know if you want to speak to it, but basically, you're not blue, you're not red, looking at, there's a quiz involved, looking at politically getting people involved, participating more at the community level of things. Each of these I think are really beautiful uses of tools to connect people, distribute information, establish relationships. I'm curious where that sits in the ecosystem of what you're doing.

Matt: We're capable of doing lots of things at the same time. Most of my energies are going into building the business at Markup AI. That means I'm actually using a lot of AI tools and getting better and better at them myself. As a thirty plus year tech CEO, I've done a lot of content creation over the years about startups, about leadership. So my blog, Only Once, has been around literally 25 years. I've published a few books. So that's more about the startup ecosystem and entrepreneurship than my current business.

And then, as you note, I have a hobby, which is current events, like lots of people. And I've done both the podcast and now this, the website you're referring to, bedrock.guide, which I just launched recently, about America and our democracy and voters, and making sure that voters are informed and that people have access to information that's outside the echo chamber sometimes. So lots of side projects. But really, my main focus is making sure that Markup AI is a strong business with great product and customers who love us. And that's where I spend most of my energy.

Caleb: Very cool. So for Markup AI, what would you say, if you had to put something at the top of the leaderboard, is the one thing Markup AI needs most right now to grow?

Matt: Well, isn't the answer to that always new customers?

Caleb: Is there anything you're seeing that you're wanting to open up bandwidth for?

Matt: Look, we just launched our platform five weeks ago, and we've actually done a pretty good job so far with demand gen. I think we have 130 or 140 new users that have created corporate accounts just in the last five weeks. So it's starting to really gather steam. But anytime you're doing something that is more what I would call category creation, which is really what we're doing, there's some work that you have to do and some passage of time that has to happen before it's clear to the world that the category needs to exist, that it does exist, and that you're the leader of it. So we try to spend most of our time making sure our product is great and delivers a lot of value for our customers. And since it's an AI product, it actually has to deliver magic. But we're also spending a lot of time thinking about the overall category and how we fit into the ecosystem, and that ecosystem has to mature. So I think we've got plenty of work to do on both fronts.

Caleb: Markup AI. Enterprise level, commercial level, any level. Content creation, being on brand, making sure compliance is looked after, ensuring you have more confidence when you're hitting that publish button. Check it out, markup.ai. Matt, thank you so much for taking the time. Really appreciate it.

Matt: Alright, good to talk to you, Caleb.